知识注入论文趋势 2023–2026

主要结论

• 知识注入论文从 2023 年的 34 篇降到 2026 年的 26 篇，热度在回落。
• CVPR 是发知识注入论文最多的会议，2023–2026 年共 38 篇，占四个会议总量的 29%。
• 交叉方向里涨得最快的是知识图谱：同时属于两个主题的论文从 2023 年的 30 篇增至 2026 年的 15 篇。
• 在「神经符号」的 5 个主题里，它的 2026 年论文数排第 5；同组增长最快的数学推理是 10 倍，知识注入是 0.8 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026DiDE: Direct Injection with Color-Texture DEcoupling for 3D Stylization
• ICML 2026SPA: A Simple but Tough-to-Beat Baseline for Knowledge Injection
• ICLR 2026NDAD: Negative-Direction Aware Decoding for Large Language Models via Controllable Hallucination Signal Injection
• CVPR 2026Distilling Balanced Knowledge from a Biased Teacher
• NeurIPS 2025From Style to Facts: Mapping the Boundaries of Knowledge Injection with Finetuning
• ICML 2025TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation
• ICLR 2025How new data permeates LLM knowledge and how to dilute it
• CVPR 2025DKC: Differentiated Knowledge Consolidation for Cloth-Hybrid Lifelong Person Re-identification
• NeurIPS 2024FEDMEKI: A Benchmark for Scaling Medical Foundation Models via Federated Knowledge Injection
• ICML 2024Enhancing Class-Imbalanced Learning with Pre-Trained Guidance through Class-Conditional Knowledge Distillation
• ICLR 2024DistillSpec: Improving Speculative Decoding via Knowledge Distillation
• CVPR 2024Aligning Logits Generatively for Principled Black-Box Knowledge Distillation
• NeurIPS 2023Improving the Knowledge Gradient Algorithm
• ICML 2023Large Language Models Struggle to Learn Long-Tail Knowledge
• ICLR 2023Noise Injection Node Regularization for Robust Learning
• CVPR 2023AttriCLIP: A Non-Incremental Learner for Incremental Knowledge Learning
• NeurIPS 2026Shared Truth: Emergent Truth Properties in Large Language Models via Heterogeneous Injection-based Transfer
• ICML 2026RedVisor: Reasoning-Aware Prompt Injection Defense via Zero-Copy KV Cache Reuse
• ICLR 2026Knowledge Distillation as Decontamination? Revisiting the “Data Laundering” Concern
• CVPR 2026Dynamic Magic Unleashing Restricted Knowledge for Lifelong Person Re-Identification

全部知识注入论文

共 131 篇，第 1 / 3 页

1. CVPR 2026ARGUS Defending Against Multimodal Indirect Prompt Injection via Steering Instruction-Following
2. CVPR 2026CamPI Physical Adversarial Examples through Camera Power Signal Injection
3. CVPR 2026Distilling Balanced Knowledge from a Biased Teacher
4. CVPR 2026Dynamic Magic Unleashing Restricted Knowledge for Lifelong Person Re-Identification
5. CVPR 2026Gated Condition Injection without Multimodal Attention Towards Controllable Linear-Attention Transformers
6. CVPR 2026Learning from Oblivion Predicting Knowledge-Overflowed Weights via Retrodiction of Forgetting
7. CVPR 2026Streamlined Knowledge Distillation
8. CVPR 2026Unlocking 3D Affordance Segmentation with 2D Semantic Knowledge
9. ICLR 2026AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM
10. ICLR 2026Inheriting Generalizable Knowledge from LLMs to Diverse Vertical Tasks
11. ICLR 2026Knowledge Distillation as Decontamination? Revisiting the “Data Laundering” Concern
12. ICLR 2026NDAD: Negative-Direction Aware Decoding for Large Language Models via Controllable Hallucination Signal Injection
13. ICLR 2026When Large Multimodal Models Confront Evolving Knowledge: Challenges and Explorations
14. ICML 2026From Teacher Pathways to Invariant Manifolds: Consensus Subspace Distillation for TSFMs
15. ICML 2026KORE: Enhancing Knowledge Injection for Large Multimodal Models via Knowledge-Oriented Controls
16. ICML 2026RedVisor: Reasoning-Aware Prompt Injection Defense via Zero-Copy KV Cache Reuse
17. ICML 2026SPA: A Simple but Tough-to-Beat Baseline for Knowledge Injection
18. NeurIPS 2026COVD: Continual Open-Vocabulary Object Detection with Novel Concept Injection
19. NeurIPS 2026DiDE: Direct Injection with Color-Texture DEcoupling for 3D Stylization
20. NeurIPS 2026LittleLearner: Language Models Under Pedagogically-Controlled Knowledge Exposure
21. NeurIPS 2026LUMOS: Tracing Parametric Knowledge from Training Data to Behavioral Outputs in LLMs
22. NeurIPS 2026On the Fragility of Latent Knowledge: Layer-wise Influence under Unlearning in Large Language Model
23. NeurIPS 2026Shared Truth: Emergent Truth Properties in Large Language Models via Heterogeneous Injection-based Transfer
24. NeurIPS 2026Similarity-Constrained Reweighting for Complex Query Answering on Knowledge Graphs
25. NeurIPS 2026VOID: Backdoor Injection through Knowledge Vacuity in Federated Unlearning
26. NeurIPS 2026You Don’t Need Aligned Representations: Knowledge Distillation via Random Prototype Spaces
27. CVPR 2025DKC: Differentiated Knowledge Consolidation for Cloth-Hybrid Lifelong Person Re-identification
28. CVPR 2025Knowledge Bridger: Towards Training-Free Missing Modality Completion
29. CVPR 2025Temporal Separation with Entropy Regularization for Knowledge Distillation in Spiking Neural Networks
30. CVPR 2025Uncertainty-Instructed Structure Injection for Generalizable HD Map Construction
31. CVPR 2025VideoWorld: Exploring Knowledge Learning from Unlabeled Videos
32. ICLR 2025Are Transformers Able to Reason by Connecting Separated Knowledge in Training Data?
33. ICLR 2025Capability Localization: Capabilities Can be Localized rather than Individual Knowledge
34. ICLR 2025ChroKnowledge: Unveiling Chronological Knowledge of Language Models in Multiple Domains
35. ICLR 2025Everything is Editable: Extend Knowledge Editing to Unstructured Data in Large Language Models
36. ICLR 2025How new data permeates LLM knowledge and how to dilute it
37. ICLR 2025KBLaM: Knowledge Base augmented Language Model
38. ICLR 2025Knowledge Localization: Mission Not Accomplished? Enter Query Localization!
39. ICLR 2025Large Scale Knowledge Washing
40. ICLR 2025Self-Updatable Large Language Models by Integrating Context into Model Parameters
41. ICML 2025ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via $\alpha$-$\beta$-Divergence
42. ICML 2025AnyEdit: Edit Any Knowledge Encoded in Language Models
43. ICML 2025Distilling the Knowledge in Data Pruning
44. ICML 2025Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding
45. ICML 2025TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation
46. ICML 2025When Data-Free Knowledge Distillation Meets Non-Transferable Teacher: Escaping Out-of-Distribution Trap is All You Need
47. NeurIPS 2025Better Language Model Inversion by Compactly Representing Next-Token Distributions
48. NeurIPS 2025Elastic Robust Unlearning of Specific Knowledge in Large Language Models
49. NeurIPS 2025From Style to Facts: Mapping the Boundaries of Knowledge Injection with Finetuning
50. NeurIPS 2025Investigating and Mitigating Catastrophic Forgetting in Medical Knowledge Injection through Internal Knowledge Augmentation Learning

常见问题

2023–2026 年四大会议一共有多少篇知识注入论文？

共 131 篇，按年份分别是 2023 年 34 篇、2024 年 39 篇、2025 年 32 篇、2026 年 26 篇。

NeurIPS 2026 有多少篇知识注入论文？

NeurIPS 2026 有 9 篇知识注入论文，上一届是 12 篇。

ICLR 2026 有多少篇知识注入论文？

ICLR 2026 有 5 篇知识注入论文，上一届是 9 篇。

ICML 2026 有多少篇知识注入论文？

ICML 2026 有 4 篇知识注入论文，上一届是 6 篇。

CVPR 2026 有多少篇知识注入论文？

CVPR 2026 有 8 篇知识注入论文，上一届是 5 篇。

哪个会议的知识注入论文最多？

CVPR，2023–2026 年共 38 篇。